Certified Legendary Thread Race for the flag, in squiggly lines

[vonn]

I don't post here week-in week-out but I appreciate it week-in-week-out*. Keep up the good work.

*no idea how the hyphen works there so put in two variations of week in, week out.**

**Now they're all covered.

 

[Seppo]

This now certainly deserves the prefix "Legendary Thread".

 

[Final Siren]

This now certainly deserves the prefix "Legendary Thread".

Certified again! This is actually the second time for this thread. It became unlegendary again at some point. Or maybe there's a timer on these things.

So thanks! It's good fun for me; I love messing around with this stuff and am really happy people enjoy it.

 

[Newmanson]

Well done Final Siren... I've looked back over the this entire thread and am hooked on this chart! The offence v defence has always been the intriguing argument in footy.

 

[Final Siren]

Final Siren You should embed my Google Spreadsheet on the first page of this thread. I'll be updating it throughout the 2015 season.

The squiggle does produce tips and predictive ladders these days, but it's primarily a visualization tool. You would expect those other models to tip better if only because that's their top priority.

The squiggle's goal is to add to your football knowledge. As far as I know, the other models you've listed are all black boxes that spit out a tip, but hide the reasoning, so you can't tell how much they're being influenced by a certain player being out injured, or the ground, or what. If you're just looking to get good tips from somewhere, that's good, but it doesn't help you understand football.

What I find interesting is how well the squiggle does at tipping despite the fact that it only knows the final scores and venues. You can know nothing about football except that, run them through a simple algorithm that anyone with Year 9 maths can understand, and produce tips every bit as good as those from football experts and advanced models.

That's pretty bizarre. You would expect someone who puts 40 hours a week into football to clearly outperform the squiggle. But they don't. Not even when armed with their own computer model, sucking in every bit of data available. They beat it by a little, or beat it sometimes, but not by anywhere near the margin that you'd expect when going from knowing almost nothing to knowing almost everything.

Which suggests to me that somewhere around 95% of what you learn about football by actually watching games, tracking injury lists, reading analysis, etc, is bunk. It doesn't help you tip any more accurately.

I find this very curious. I'm not sure why it is. There is probably a lot of "common knowledge" that's actually wrong. A lot of things people think are important but actually aren't, and vice versa.

 

[KissKiss]

I've brought this thread/information up with a few friends who were talking about the old statement "defence wins titles", and one thing someone kept arguing is that offence seems to win more in AFL but this is because the rules keep being changed to stop defensive tactics

 

[Final Siren]

I've brought this thread/information up with a few friends who were talking about the old statement "defence wins titles", and one thing someone kept arguing is that offence seems to win more in AFL but this is because the rules keep being changed to stop defensive tactics

Ha, does that even matter, though? If the rules gave nerdy bald guys free kicks, I'd be a superstar, but they don't, so I'm not.

 

[KissKiss]

Ha, does that even matter, though? If the rules gave nerdy bald guys free kicks, I'd be a superstar, but they don't, so I'm not.

No idea, but he has two major things he loves talking about. One is that defence wins (hence I started showing him this data) unfortunately that lead into his other favourite topic in that the AFL keeps ruining the sport by changing the rules

 

[Final Siren]

I love this. I sort of understand it, but i still love it.

Just a question for you Final Siren, does this give any indication on who it thinks will win the flag? Does it show anything other than the attack and defence of a team? I can understand it all but i just am having trouble interpreting it all into information.

It does in that premiers tend to wind up in the same kind of area on the chart. And they tend to be closer to that area than anyone else. You can get a sense for this by browsing past years here.

The current hot topic is whether teams like Sydney and Fremantle should try to be more attacking. Because despite 2012 and 2005, the right-side of the chart (high-defence, moderate-attack) has been a bit of a graveyard for good teams. They finish top 4 but they miss a lot of flags.

 

[clogged]

The squiggle does produce tips and predictive ladders these days, but it's primarily a visualization tool. You would expect those other models to tip better if only because that's their top priority.

The squiggle's goal is to add to your football knowledge. As far as I know, the other models you've listed are all black boxes that spit out a tip, but hide the reasoning, so you can't tell how much they're being influenced by a certain player being out injured, or the ground, or what. If you're just looking to get good tips from somewhere, that's good, but it doesn't help you understand football.

What I find interesting is how well the squiggle does at tipping despite the fact that it only knows the final scores and venues. You can know nothing about football except that, run them through a simple algorithm that anyone with Year 9 maths can understand, and produce tips every bit as good as those from football experts and advanced models.

That's pretty bizarre. You would expect someone who puts 40 hours a week into football to clearly outperform the squiggle. But they don't. Not even when armed with their own computer model, sucking in every bit of data available. They beat it by a little, or beat it sometimes, but not by anywhere near the margin that you'd expect when going from knowing almost nothing to knowing almost everything.

Which suggests to me that somewhere around 95% of what you learn about football by actually watching games, tracking injury lists, reading analysis, etc, is bunk. It doesn't help you tip any more accurately.

I find this very curious. I'm not sure why it is. There is probably a lot of "common knowledge" that's actually wrong. A lot of things people think are important but actually aren't, and vice versa.

I'd say the squiggle self-corrects in the case of injuries, players available and the natural declines and rises of clubs. Take Geelong vs Freo on the weekend. The squiggle tipped Geelong, but only by a point. We're essentially splitting hairs there, so let's assume that the squiggle predicted Freo to win, by a point. But Freo won by 44 points. So squiggle gets the tip right and has the margin built into it's future calculations of just how Geelong and Freo will go.

So while it might get wrong occasionally (it tipped Geelong) there are occasions where it gets it right and wrong at the same time. The tip is right, the margin is wrong, but the margin feeds into the model to make better predictions about both clubs next time.

The squiggle is predicting Hawthorn to beat the Bulldogs by ~10 goals, but if instead the Hawks only get up by say 20 points, then the model is still correct in terms of wins and losses, and knows enough about the changing fortunes of both clubs to better predict their results in the future.

 

[AlibiMonday]

The squiggle does produce tips and predictive ladders these days, but it's primarily a visualization tool. You would expect those other models to tip better if only because that's their top priority.

The squiggle's goal is to add to your football knowledge. As far as I know, the other models you've listed are all black boxes that spit out a tip, but hide the reasoning, so you can't tell how much they're being influenced by a certain player being out injured, or the ground, or what. If you're just looking to get good tips from somewhere, that's good, but it doesn't help you understand football.

What I find interesting is how well the squiggle does at tipping despite the fact that it only knows the final scores and venues. You can know nothing about football except that, run them through a simple algorithm that anyone with Year 9 maths can understand, and produce tips every bit as good as those from football experts and advanced models.

That's pretty bizarre. You would expect someone who puts 40 hours a week into football to clearly outperform the squiggle. But they don't. Not even when armed with their own computer model, sucking in every bit of data available. They beat it by a little, or beat it sometimes, but not by anywhere near the margin that you'd expect when going from knowing almost nothing to knowing almost everything.

Which suggests to me that somewhere around 95% of what you learn about football by actually watching games, tracking injury lists, reading analysis, etc, is bunk. It doesn't help you tip any more accurately.

I find this very curious. I'm not sure why it is. There is probably a lot of "common knowledge" that's actually wrong. A lot of things people think are important but actually aren't, and vice versa.

I think ISTATE-91:12 is a beautiful example of the power of simplicity. But you can't undersell the fact that you've rigourlessly tested possibly hundreds of permutations over years worth of data - much in the way machine learning works. To me, that's the killer. I'd venture to say that the other black box models are being continually tweaked, without being run again and again based on variables they've used. They're at the real risk of overfitting the data to very recent results which can be terrible for inference.

I think BF is most fortunate that you've found a sweet spot between football, statistics and some incredible web skills. I know I feel fortunate and hope your energy continues.

BTW - not a bad writer too

 

[clogged]

On the Geelong situation, I'd imagine that last year the squiggle tipped Geelong games correctly ~75% of the time. And Geelong finished top four last year. So given the squiggle mostly tipped Geelong correctly and they finished in near the top, a 'dumb' model would surely tip them to finish in the same place this year.

But the squiggle didn't, at the start of the year it predicted Geelong to miss the finals altogether. And anyone watching the way Geelong won many of its games last year would have told you that. Just like a person, the squiggle doesn't rate 'unconvincing' wins. If you have too many of those, and it will pop you out of the finals.

 

[Glenno23]

Subsquiggled

 

[The Swans Blog]

I'd say the squiggle self-corrects in the case of injuries, players available and the natural declines and rises of clubs. Take Geelong vs Freo on the weekend. The squiggle tipped Geelong, but only by a point. We're essentially splitting hairs there, so let's assume that the squiggle predicted Freo to win, by a point. But Freo won by 44 points. So squiggle gets the tip right and has the margin built into it's future calculations of just how Geelong and Freo will go.

Just about every model predicted that Geelong were going to win. Roby's predicted Geelong before the Motlop change, and everyone I knew was going to tip Geelong, heck even I did at first. Of all the models I'm tracking, it seems that his is the only one that puts value on players.

So while it might get wrong occasionally (it tipped Geelong) there are occasions where it gets it right and wrong at the same time. The tip is right, the margin is wrong, but the margin feeds into the model to make better predictions about both clubs next time.

The squiggle is predicting Hawthorn to beat the Bulldogs by ~10 goals, but if instead the Hawks only get up by say 20 points, then the model is still correct in terms of wins and losses, and knows enough about the changing fortunes of both clubs to better predict their results in the future.

Squiggle typically has lower error compared to power rankings whereas PR might pick 1 or 2 correctly where Squiggles doesn't, because of injuries or omissions. What this suggests is that there's not enough value in rankings players because it doesn't take into account subjective performance measures, such as performance. Who would have thought that Sydney and Hawthorn would have performed so poorly against Essendon? It isn't that Essendon is a vastly superior team - they aren't, it's just that those teams had very low motivation.

The same can be said for Swans who thrashed Port, or Saints last round, or the Dogs the last 2 rounds.

 

[Rotayjay]

'The squiggle loves a good thrashing'. Hahaha, I love this thread. Deserves the Legendary certification.

 

[The Swans Blog]

Final Siren so far Squiggles leads the way in season 2015 for predicting results as well as prediction error.
http://www.bigfooty.com/forum/threads/afl-prediction-models-round-3.1094371/

In my comparison of the 5 main prediction models, Squiggles won rounds 2 and 3. Only Footy Forecaster has picked 1 closer result and Massey Ratings has slightly less error.

Excellent picks so far; may it live a long, successful tipping life!

 

[Final Siren]

Round 3, 2015​

Animated!

Good week for: Fremantle, who exceeded squiggle expectations even after the Eagles piled on a bunch of junk time goals. It's been a pretty solid three weeks for the Dockers, who are threatening to develop a pretty good trend.

Bad week for: Geelong. The Cats had an injury shocker and won plaudits for their fighting spirit. But that shouldn't distract from the fact that they were playing the freaking Gold Coast at Kardinia Park. The Suns have been terrible this year, losing to Melbourne in Round 1 and St. Kilda (at home) in Round 2, and the Cats should have been able to account for them more comfortably than that.

Additionally, four clubs that are Geelong's closest competition for a finals spot all did pretty well this week: Richmond, Essendon, North Melbourne, and West Coast.

A rotten week for the Lions, too, who have also trended badly in all three rounds. In fact, the squiggle predictor sez:

The squiggle's top 4 hasn't changed since Round 1, with four teams from four states. And if you had to pick a change, you'd probably replace Adelaide with Port Adelaide, which would maintain that.

So home finals will be worth a premium, especially those top two spots. The squiggle awards a 12-point advantage when playing an interstate side, which becomes a 24-point difference between playing an interstate side at home vs playing them away. Four goals is a massive difference, and it definitely seems to exist in the first weeks of finals, when home teams don't just win more often, but win more often than home teams normally do. That makes every game a must-win for the top teams.

Also on the horizon: a pretty amazing run for Adelaide. The Crows' next five games are against the Bulldogs, Power, Suns, Saints, and the Giants. And it doesn't get too much tougher for them after that. But their butt-end of the season is hard.

 

[The Swans Blog]

Going to be a lot of offended when the Crows are basically handed a preliminary final - again.

 

[Suntory Boss Coffee]

Going to be a lot of offended when the Crows are basically handed a preliminary final - again.

Then for them to actually play well but let the umpires win it for the Hawks again ahahha

 

[clogged]

Melbourne big improvers simply by losing in a respectable way. From Round 2:

and Round 3:

 

[Roo Lives Again]

Going to be a lot of offended when the Crows are basically handed a preliminary final - again.

You're an absolute troll

 

[The Swans Blog]

You're an absolute troll

There are undeniable similarities between Adelaide's 2012 and 2015 seasons. Favorable home draw, double-ups against significantly weaker opposition and most difficult games at home. Its probable that Adelaide will finish 4th and probably play the qualifying final against Sydney, again.

 

[ROYAL EAGLE]

Going to be a lot of offended when the Crows are basically handed a preliminary final - again.

What do you mean handed a prelim ?

 

[djpaulydemon]

15th with 6 wins. One can only dream!

 

[Roo Lives Again]

There are undeniable similarities between Adelaide's 2012 and 2015 seasons. Favorable home draw, double-ups against significantly weaker opposition and most difficult games at home. Its probable that Adelaide will finish 4th and probably play the qualifying final against Sydney, again.

Lets get into the debate about the AFL handing things to certain clubs on a platter. You'll lose that one by a country mile.